Why This Matters

Asking 'list the bugs' produces prose; asking 'output as a JSON array of {severity, issue, fix}' produces parseable data. Output shaping saves a follow-up turn and unlocks automation (feeding Copilot's output to a script).

Concept Explained

Output shaping means specifying the response format in the prompt. Common formats: bullet list, numbered list, table, JSON, diff, markdown code block. The format shapes what the model produces, not just how it looks.

How It Works

Format instructions bias the model toward structured output. Asking for JSON activates JSON-producing patterns; asking for a table activates table-markdown patterns. Combine with a schema for maximum control: 'output as JSON matching this TypeScript type: {severity: 'high'|'med'|'low', issue: string, fix: string}'.

Step-by-Step Tutorial

1. Pick a format

2. Specify in prompt

3. Add schema for JSON

4. Combine with role

5. Validate

Real-World Example

A team wanted to feed Copilot's bug review into their issue tracker. They asked: 'Review this function. Output as JSON array of {severity, title, description, suggested_fix}. No prose.' Copilot produced clean JSON. They wrote a 10-line script to file issues automatically. Saved ~30 minutes per review.

Example Prompts / Commands / Code

# Bullet list:
List the top 3 performance issues. Output as a bullet list, no prose.

# Table:
Compare these 3 libraries. Output as a markdown table with columns: library, stars, last_release, license.

# JSON with schema:
Review this function for security issues. Output as JSON matching this TypeScript type:
type Issue = { severity: 'high'|'med'|'low'; issue: string; fix: string };
Return Issue[].

# Diff:
Refactor this function. Output as a unified diff, no prose around it.

# Code block only:
Implement the function. Output as a single TypeScript code block, no explanation.

Common Mistakes

• Asking for JSON without a schema — produces inconsistent shapes.
• Mixing prose with structured output — defeats the purpose.
• Forgetting to validate JSON — silent parse failures downstream.
• Using tables for >5 columns — hard to read.

Best Practices

• Match format to use: JSON for automation, table for comparison, diff for edits.
• For JSON, always specify a schema.
• End with 'no prose' or 'no explanation' if you want clean output.
• Validate structured output before consuming.

Troubleshooting

Practical Exercise

Frequently Asked Questions

Further Reading

• Structured outputs with Copilot

Official References

• Prompt engineering

Related lessons: BE-17, BE-19, IN-04